Recently I did a construction site. Shot it a 1in/pixel GSD with a Phantom 4 Pro. Frontlap 75%, Sidelap 75%. I gave the orthophoto (GeoTIFF) to the Land Suveying company that surveyed the property and the photo wouldn’t line-up right when they brought it into Carlson Survey. It was fairly close but a lot of the building edges in the photo were quite a bit off from the survey drawing, a few feet off.
My understanding that is that relative accuracy should be 2-3 times the GSD. So if I shot at 1in/pixel, then things should line-up within 3 inches, no?
We didn’t use ground control points because this was actually an inspection job for the lender to view the status of the project and the Surveying Company that was working with the lender neglected to tell us upfront they wanted to overlay it on an existing survey map. However I’m trying to figure out what the cause of the inaccuracy would be as I would think 75/75 overlap and 1in/pixel GSD should produce pretty good relative measurement accuracy. Yes, the geolocation coordinates in the GeoTiff would be off since we had not ground control points, but I would think you should be able to manually align it.
We were joking that maybe the builder didn’t build the buildings correctly! But we are assuming they did. The site was only about 5 acres.
For correct positioning of an ortophoto you MUST have GCP. Remember the GPS in the Phantom can have an offset of 10m…
And even with CGP remember also that the ortophoto are images arranged together so high buildings or structures can appear a little bit off…
I’m not talking about the accuracy of the geolocation coordinates in the GeoTiff. As I mentioned I’ve read in multiple places the accuracy of the relative measurements in an orthophoto are typically 2-3 times the GSD. So without ground control points, that means if the photos were all shot with 2in/pixel GSD, if I take a measurement of a distance in the orthophoto, the measurement should be accurate to within 2x3 = 6 inches. It is my understanding that photogrammetry involves identifying the same visual points in multiple photos and that the GPS coordinates from the camera are only used to determine which photos are nearby each other because as you point out the GPS isn’t accurate enough.
So what I’m saying is yes, the lat/long coordinates could be off, but you can manually position the orthophoto on a map base in the correct location and thereby the photo theoretically would only be off by 6 inches when comparing specific locations of objects in the photo to more precise positions in like a land survey base drawing.
However, we saw variances of 1-5 feet using an orthophoto with 1inch/pixel GSD. This is completely contradictory to the information I’ve seen published about GSD and *relative* measurement accuracy.
That’s why you should… or put some GCP or do a Scale constraint… to put the results to the wright scale.
When you don’t use any of this the software make de model without any dimensions constraint it just calculates and stick images together. If you use GCP the model its “obligated” to be on scale. If you don’t have any GPS you can use a tool in pix4D which is new scale constraint
All this because the GPS on the drone it’s not accurate enough. You should put the model in scale before take measurements.
The only way to put the model in scale it’s to give accurate GCP or accurate scale constraint
As mentioned by @Jorge , to reach high accuracy, it is true we always recommend including ground control points (GCPs) in projects.
When it is not possible, make sure there is a feature (or more) in your project of which you know the length, to (a) check the reconstruction and if necessary (b) to use as a scale constraint. In a few words, you provide the software with the real measured size of some feature (i.e. length of a building) and, after you reoptimize, it will scale the project accordingly, which should increase the relative accuracy. To learn more, check out our How to scale a project article.
Regarding the relative accuracy of a project, an error 1-3 times the ground sampling distance (GSD) is expected. However, it depends on many factors related to the quality of the reconstruction. To learn more about how we define the relative and absolute accuracy of a project, check out Accuracy of Pix4Dmapper Outputs.
Keep in mind that this error is not evenly distributed in the project, as the relative accuracy depends on the overlap between images, the visual content of the images, or the GSD, which are all parameters that can differ between images. As a consequence, some points will be more accurate that others.
Additionally, the scale is the limiting factor, as some problems are magnified at a higher resolution, such as perspective, vibrations of the camera, etc.
@Jorge and @Jeff
Note that the orthomosaic is not generated by photo stitching, but by orthorectification, which removes the perspective of the camera. Each 3D point of the project is reconstructed from the pictures it appears in. For more information about this, see Photo Stitching vs Orthomosaic Generation.
Rhéa thanks for the enlightening of the orthomosaic. Actually I did know the process took in account the point cloud, but I thought (wrongly) that was a more simple process. That one described in the post Photo stitching vs. Orthomosaic generation its clear and yes it’s a much precise method.
Thanks a lot. Today I know more than yesterday…
Wow, there is a lot of misconception in this thread so let me summarize a few points from the dozens and dozens of tests I have done for both horizontal and vertical 3D modeling.
#1 is the 1X to 3X GSD accuracy is in the point cloud, not the orthomosaic. Sharp changes in elevation just don’t show good accuracy in an orthomosaic. In fact I never bother to make the orthomosaic as the point cloud is the truest representation.
#2 a relative accuracy very near to 1X GSD can be achieved without any GCP or scale constraint. Most people don’t believe this to be true but I have the process and repeatable data that proves the capability of Pix4D. I can get sub-mm relative accuracy on top of a cell tower with only using GPS, not even RTK.
#3 all of the above is 100% dependent on how the pictures are taken. In this construction site with at least one building, nadir images just won’t cut it for high accuracy so be sure to use oblique camera angles. I go all the way to 45 degrees but that isn’t always necessary.
So if the end goal is to provide an overhead photo of a property along with a measurement tool for measuring distances between objects viewed in the photo, what good is a point cloud going to do you OUTSIDE Pix4D? Maybe end user is a construction company wanting to measure distances between buildings, etc. They’re not "sophisticated’ from a technical standpoint. They need a free, easy (preferably online tool) to view the orthomosaic photo and measure the distances. Not everyone lives in Pix4D you know. So explain how this example end user that isn’t using Pix4D is going to use your point cloud to measure distances look for objects in a photo, OUTSIDE Pix4D. As far as I know, you give them a GeoTiff. That’s NOT a point cloud.
There is more than one online service to use and measure point clouds and even more if you convert it to a mesh but that loses some detail. Personally I prefer Pix4D for measuring since you get the associated pictures to automatically display at the same time…and at $350/month it is very low cost and very easy to use. But any of the other point cloud viewers would work too.